Crowd size estimation is an important issue for event organizers, law enforcement, urban planners, and news media, among others, who want accurate estimates of crowd size at a specific time and location. For example, organizers may want accurate estimates of turnout to demonstrate support for their cause; law enforcement, which needs to optimally allocate critical resources to secure an event, may want accurate crowd size estimates to ensure that resources are allocated in the best and most cost effective manner; urban planners may use crowds size estimates to design improved solutions for handling crowds in public spaces such as parks, sports venues, and the like; and news media may use crowd size estimates to report the attendance at an event.
Several methods for crowd estimation have been proposed with varying degrees of success. One such method uses aerial photographic analysis to estimate crowd sizes In the aerial photographic analysis method, a fixed-wing aircraft is used to take aerial photographs at an altitude of 2,000 feet or less. Photographs of the area are taken in strips using a digital camera with about sixty-percent overlap between successive pictures to allow stereoscopic viewing. An image resolution of about one foot per pixel is typically used. The photographs are then loaded in an image processing program and co-registered with a one-meter-resolution United States Geological Survey ortho-photo map—a perspective-corrected collage of aerial shots of the area with a uniform scale. A grid is then superimposed on the image. Units are classified by the apparent density of people per unit. A cross, dot or other marking is placed on each individual's head or shadow point. Each marking is counted or, if necessary, estimated to determine the number of people in each grid unit. An error is then calculated based upon the number of grid units divided by the degree of uncertainty about how many people each grid unit contains.
Aerial photographic analysis works under certain assumptions and conditions. Flying over the crowd during peak times requires an initial estimate of when that peak time occurs since photography is, by definition, a snapshot in time. The methods presented herein below provide continuous estimation for the duration of an event. Crowd estimation over time to find dynamic crowd size estimates may be critical for first responders and law enforcement to assess a volatile situation before it reaches a flash point. For example, a rapidly growing mob could raise an alert to local law enforcement and event organizers.
Aerial photographic analysis also requires conditions that permit the use of aircraft flying at a certain altitude. In contrast, the disclosed methods allow a system to remotely analyze data collected from cell phones of the participants using wireless performance metrics. Analysis of the collected data can be done after the fact in situations where the event was not advertised or promoted and aerial photography was not permitted or planned.
Another pitfall of aerial photographic analysis occurs if the event is geographically dispersed. In such situations, the cost of aerial analysis increases along with the costs of coordinating the measurement at multiple event sites. Moreover, this method requires pre-knowledge of the event sites and does not permit after-the-fact analysis of crowd sizes. The disclosed method is agnostic of the number of locations or pre-knowledge since the disclosed data collection method can be continuous. In some cases, enhanced metric collection may be needed and can be activated remotely as needed. Thus, the disclosed method addresses a significant gap in cases where event sites are geographically dispersed.
Crowd estimation using aerial photographic analysis requires some degree of technical skill and, if done incorrectly (e.g., low resolution, blurry, or otherwise unusable photographs), will adversely impact the crowd size estimates. In some cases, such as events at night or events held in restricted spaces, aerial photography may not be possible. The disclosed methods can be used to complement aerial photography analysis in addition to addressing cases in which aerial photography methods is not preferred or will not work due to low light levels, dispersed crowds, vegetation, terrain, or participants being inside building or temporary structures such as tents, port-a-potties, and the like.
The crowd size estimates provided by the disclosed method can reduce the errors in aerial estimates by improving the degree of uncertainty about how many people attended a particular event. The disclosed method provides another dimension to event attendance.
Other methods for crowd estimation exist and have their own constraints. For example, crowd estimation by ground-level surveillance video, also called detector-based analysis, is only accurate on a small scale. This method requires extensive image analysis algorithms and is not scalable to large crowds. This method requires management of complex issues such as motion detection, clustering, pixel change analysis, vanishing points, fractal analysis, perspective views, and video quality.
Another crowd estimation method uses measurements of dynamic flows across a monitored boundary. Several entry, exit, and intermediate points on a parade route are monitored and the rate of movement of people across the boundary line is measured. These are extrapolated over time to estimate a crowd size in attendance at the parade.
Yet another crowd estimation method uses crowd density and location segmentation. This method maps an event space into a series of segments and, based upon the dimensions of the segment, computes a packing density and, therefore, an attendance number for the event space. This method has been used to estimate the crowd size at New Year's Eve events at Times Square in New York City, N.Y.
Another crowd estimation method uses the amount of trash produced by attendees. This method arose from the days when ticker tape parades yielded measurable trash volumes. Trash production by modern era crowds is comparatively small. Crowd estimation by trash volume is no longer viable due to many error factors. Similar modern methods that exist monitor port-a-potty use and extrapolate from there to determine crowd size.